Objective Estimation of the 24-Hour Probability of Tropical Cyclone Formation

The algorithm developed in this study provides a probabilistic, short-term forecast of tropical cyclone formation. Since TC formation is an extremely rare event, the typical statistical verification methods that would be applied to a probabilistic forecast have biases that make many of them insufficient for estimating skill. For example, consider that for a probability threshold p = 0.4% chosen for the Atlantic basin, the corresponding hit rate and false alarm rate are HR = 65.2% and FAR = 2.3%. These values are not far off from those reported in HH03and PL86, and at face value appear to suggest a good forecast. However, with 428,553 NTCG cases in the Atlantic dataset this equates to more than 9,800 false alarms. In the generalized framework of this analysis TC genesis is an extremely rare occurrence (e.g., for the Atlantic (A+B)/(C+D) ~ 0.0016 << 1) which makes is difficult to compare existing results that rely on restricted initial sample sets. As such, we will attempt to define an acceptable methodology for evaluating the skill of our TC formation prediction scheme.

Although standard verification measures for probabilistic forecasts are not suitable for use with extremely rare events, there are some variations of these metrics that compare the performance of a forecast to that of a reference forecast to determine skill. The best available reference forecasts in this case are the null forecast (assuming p = 0% everywhere) and the climatological formation probability, CPROB. The results of each verification method are given for both the dependent (1995 – 2005) and independent (2006) datasets.
a. Brier Skill Score

The first verification diagnostic used was the Brier Skill Score, BSS = 1 – (BS/BS_ref), where BS is the Brier Score of the algorithm and BS_ref is a reference Brier Score (Murphy 1973). The Brier Score, as defined by Australian Bureau of Meteorology (cited 2007) is, BS = 1/N Σ i=1,n (p_i-o_i)², where p_i and o_i are the predicted and observed probability for case i, respectively. In this case, the reference forecast is defined as 1) the null hypothesis and 2) the climatological TC formation probability (CPROB). The BSS measures the improvement of a probabilistic forecast over the reference forecast and can range from -∞ to 1, with 0 indicating no skill with respect to the reference forecast and 1 being a perfect score. The BSS for the dependent years, using the null hypothesis for BS_ref, are 0.016, 0.032 and 0.029 for the Atlantic, E. Pacific and W. Pacific basins, respectively. The corresponding independent year BSS are 0.011, 0.035, and 0.020. When the climatological formation probability is used for BS_ref,, the Brier Skill Scores for the dependent years are 0.012, 0.024 and 0.025 for the Atlantic, E. Pacific and W. Pacific basins, respectively. The corresponding independent year BSS are 0.009, 0.027, and 0.018. Although the BSS values are relatively small, they are all positive which indicates that the algorithm provides a more skillful forecast than both the null hypothesis and climatology alone.

Another gauge of forecast skill that can be used for this scheme is the Relative Operating Characteristic (ROC). This value is determined by plotting the hit rate versus the false alarm rate using a set of increasing probability thresholds to make the yes/no decision (Mason and Graham 1999; Australian Bureau of Meteorology, cited 2007). The area under the ROC curve, which ranges between 0 and 1, is the ROC score and provides information regarding the ability of a forecast to discriminate between events and non-events. A score of 0.5 or less indicates no skill and a perfect score is 1. The ROC scores for the Atlantic, E. Pacific and W. Pacific basins are 0.81, 0.88, and 0.90, respectively (Figure 5). However, as discussed at the beginning of this section, these scores alone are not necessarily representative of the overall forecast skill of the algorithm, due to known problems with using the ROC for extremely rare events (Stephenson 2004). Hence, for the purpose of this study we will define a ROC Skill Score (RSS) as RSS = (ROC_alg – ROC_clim)/ROC_alg, where ROC_alg and ROC_clim are the ROC for the algorithm forecast and climatological forecast, respectively. Like the Brier Skill Score, RSS can range from -∞ to 1, with any value RSS > 0 indicating forecast skill with respect to climatology. The RSS for the Atlantic, E. Pacific and W. Pacific basins are 0.21, 0.01 and 0.22, respectively. These values indicate that the algorithm forecast is skillful with respect to climatology in all three basins, although skill is only marginally better than climatology in the E. Pacific. The relatively large values of climatological formation probability (Fig. 3) and the heavy weighting of CPROB as an input parameter for the LDA (Table 4) in the E. Pacific basin suggest that climatology is a strong predictor of TC formation probability in that basin. The E. Pacific basin has the highest frequency of TCG per unit area of any region worldwide (Elsberry et al. 1987), which further supports the idea that TCG formation varies less from year to year there than in other basins. This attribute makes climatology a good predictor of TCG in the E. Pacific and hence a tough forecast to beat, regardless of the methodology used.

The algorithm developed in this study has been implemented as the NESDIS Tropical Cyclone Formation Probability (TCFP) product, which is currently running in real time at http://www.ssd.noaa.gov/PS/TROP/genesis.html. The TCFP product website includes a domain-wide overview plot indicating regions of elevated TC formation probability along with a corresponding enhanced water vapor loop. Each of the three basins has its own web page that displays color contour xy-plots of real-time, climatological, and anomaly formation probability and predictor values (Fig. 6).

Each basin has been divided into a set of sub-basins as shown in Fig. 7. To provide continuity over time, both the product-estimated and climatological formation probabilities (input parameter values) are summed (averaged) over each sub-basin and displayed as time series plots. These time series are useful for identifying sub-basin scale variations in input parameter values and the corresponding changes in TC formation probability. The time series plot of 24-hour predicted TC formation probability over the Eastern Pacific (Fig. 7) sub-basin is shown in Fig. 8. The red open circles in Fig. 8 represent the times within 24 hours prior to TC formation. As Fig. 8 shows, most of the red open circles are coincident with peaks in sub-basin summation TC formation probability. In fact, for the Eastern Pacific sub-basin 81.5% of the TCG points occur at times when the sub-basin formation probability is greater than its climatological value and over half of the TCG points are coincident with a peak in the TC formation probability. It turns out that this result is robust over most sub-basins, with 80.3% of the total TCG points occur at times when the sub-basin formation probability is greater than its climatological value. The fine resolution of the analysis grid introduces the problem of TCs moving in and out of grid boxes during the forecast period, which leads to a negative bias in algorithm hit rates. Using the sub-basin summation of formation probability reduces this problem and gives a more representative view of formation probability when a region of increased TC formation probability spans a large region, such as in an active phase of the monsoon trough or an elongated easterly wave. These time series may prove as useful as the sub-region TC formation probabilities to forecasters.

After examining sub-basin averages and sums, it seems prudent to examine these probabilities on the basin scale. To do so, the product probabilities were summed over the entirety of each basin for each dependent and independent year and compared to the observed number of TC formations in that basin. The results from this computation are plotted in Fig. 9. Although the probabilities have been bias-corrected so that the cumulative predicted probabilities equal the total number of TC observations over the dependent dataset, the difference between predicted and actual number of TC for individual years is apparent in Fig. 9. Most of the data points in Figure 9 are relatively close to the x=y line (dashed), indicating that the TCFP product is performing well on the annual, basin-wide scale.

A product for predicting the 24-hour probability of TC formation within each 5° x 5° latitude/longitude sub-region in the Atlantic, E. Pacific and W. Pacific tropical basins was developed. This product uses both environmental parameters from ATCF best tracks and NCEP GFS data fields as well as convective parameters from the GOES-E, GOES-W and MTSAT-1R channel-3 water vapor imagery. The predictive algorithm was developed using a two-step process that involves 1) screening out data points where TC formation is highly unlikely and 2) linear discriminant analysis. The resultant discriminant function values are interpolated to 24-hour TC formation probabilities.

Verification of the TCFP product, using both the Brier Skill Score and Relative Operating Characteristic, showed that the product forecast possesses skill with respect to both the null hypothesis and climatology. Although individual product probability values are generally no greater than 10-15%, they are a vast improvement on climatological values, which are on the order of 0.1%. In addition, the mean algorithm-derived 24-hour TC formation probabilities for TCG cases is 1.5, 3.0 and 2.6% for the Atlantic, E. Pacific and W. Pacific basins respectively while the corresponding means for NTCG cases are 0.04, 0.03 and 0.07%. The differences between the group means were found to be significant at the 99% confidence level using the Student’s T statistic.

b. Future Work

Several different approaches for improving the TCFP product are currently being considered. The first involves extending the analysis domain to include the Indian Ocean and Southern Hemisphere tropical cyclone basins, essentially making the product global. The real-time, objective TC formation guidance supplied by this product could be of great use to meteorological agencies in these regions, including JTWC, National Weather Service (NWS) offices in Pago Pago, American Samoa, Fiji TC Regional Meteorological Specialized Center (RMSC), La Reunion RMSC, and the Perth, Darwin and Brisbane TC Warning Centres.

Another improvement involves identifying additional environmental parameters to add to the algorithm. Two such parameters being considered are the positions and Dvorak intensities for invest tropical systems supplied by the NCEP/TPC Tropical Analysis and Forecast Branch. Lastly, the current TCFP product determines the probability of TC formation within 24 hours, which is a short forecast period. Future research will investigate ways in which the forecast period can be extended to 48 hours and beyond. This may include use of the GFS forecast fields in addition to the analyses. Also, Frank and Roundy (2006) demonstrated that TC formation is strongly related to several types of atmospheric waves and identified corresponding phase relationships. Given the fact that convective anomalies associated with these waves are detectable up to a month in advance of TC formation, the statistical TC formation forecasts developed by Frank and Roundy (2006) introduce the idea of using upstream convective parameters to extend the TCFP product forecast period. Work is currently underway to collect and analyze global water vapor imagery for potential upstream wave-related convective signatures that can be incorporated into the current algorithm.
Acknowledgments

This research was partially supported by the GOES PSDI project. The authors would like to thank Lixion Avila, Ed Rappaport, Chris Landsea and Edward Fukada for providing motivation and guidance in developing this product, Bernadette Connell for supplying GOES-West water vapor imagery, and Ray Zehr for supplying GMS-05 and MTSAT-1R water vapor imagery from the Tropical RAMSDIS archives. The views, opinions, and findings in this report are those of the author and should not be construed as an official NOAA or U. S. government position, policy, or decision.

Latitude can have an important impact on TC formation, particularly in terms of limiting its likelihood. TC formation is rare at low latitudes (Fig. 2), where the Coriolis parameter (and hence planetary vorticity) is nearly zero and surface winds (and hence pressure gradients) are generally weak (Gray 1975, 1979). For this study, the latitude (LAT) is defined at the center of each sub-region.

Energy fluxes between a tropical cyclone and the ocean surface play a crucial role in TC formation and intensification (Malkus and Riehl 1960, Leipper 1967, Gray 1975). For this reason, it is highly unlikely for a tropical cyclone to form over land (Fig. 2). For these reasons, the percent of each sub-region over land (PLAND) was included as a parameter.

An existing TC introduces storm-related vertical shear, change the local atmospheric moisture content and cause upwelling that decreases SST, all of which result in an environment that is unfavorable for new TC development. Analysis of the 1949-2005 best tracks confirms that in the last 57 years, no TC has ever formed within 400 km of an existing TC, justifying the inclusion of this distance as a parameter. Distance to an existing TC (DSTRM) is calculated from the ATCF best track positions for each analysis time. Since the effects of an existing TC are localized, variations in distances beyond 1000 km are not expected to have any predictive quality and hence all values of DSTRM greater than 1000 km are set equal to 1000 km.

Shapiro and Goldenberg (1998) demonstrated that sea surface temperatures (SST) have a direct effect on enhancing TC formation. The exact threshold value for SSTs that support TCG is still unknown, however Palmén (1948) suggested an SST of 26 °C was needed to support tropical cyclone formation. The 24-hour values of CSST were derived by linearly interpolating between monthly SST values.

In a study of west Pacific typhoons, Riehl (1948) found that TC formation can be inhibited by strong vertical shear of the horizontal wind. In addition, several studies have suggested that some weak amount of vertical shear is necessary for TC development (e.g., McBride and Zehr 1981; Bracken and Bosart 2000). This relationship between vertical shear and TC formation was recognized by Gray (1968) and DeMaria et al. (2001), and was included as dynamical parameters in their respective TC genesis parameters. For this study, the magnitude of the vertical shear of the zonal wind (VSHEAR) is calculated from the GFS analysis horizontal wind fields for each sub-region as,



where _ and _ are the zonal and meridional winds , respectively, averaged over a 12° x 12° latitude/longitude GFS sub-grid that is centered on the analysis sub-region.

Gray (1968, 1975, 1979) noted tropical cyclones tend to form in regions of anomalous positive vorticity. The observational studies of Frank and Clark (1980), McBride and Zehr (1981) and others confirm the importance of low-level positive circulation and vorticity in determining formation potential. This increase in low-level positive circulation enhances heat and moisture fluxes from the ocean surface and, in the presence of deep convective bursts, leads to the development of the low-level, warm core TC vortex (Hendricks et al. 2004; Montgomery et al. 2006, Hendricks and Montgomery 2006). The 850 hPa circulation (CIRC) for each sub-region in the algorithm domain is calculated from the GFS wind fields. An average is taken over an 8° x 8° sub-grid of the GFS data field that is centered on the algorithm sub-region.

Vertical instability is needed to support the deep convective activity necessary for TC formation. The vertical instability parameter from DeMaria et al. (2001), which was developed using the parcel calculation formulation described in Ooyama (1990), is used in this study. Vertical instability (THDEV) is calculated for each sub-region using the equation;

where wt(p) is the pressure weighting and _ and _ are the equivalent potential temperature and saturated equivalent potential temperature averaged over a 8° x 8° sub-grid of the GFS data field.

Several studies of TC formation climatology suggest that TC formation is preferred in regions of large-scale enhanced low-level convergence. In particular, TC formation is often associated with enhanced convergence zones associated with equatorial Rossby waves (Molinari 2007), the Madden-Julian Oscillation (Maloney and Hartmann 2001), and the monsoon trough (Briegel and Frank 1997). Based on these findings, the 850 hPa horizontal divergence was included as a parameter. The 850 hPa divergence (HDIV) for each sub-region in the algorithm domain is calculated from the GFS wind fields;

where _ and _ are the values _ and _, respectively, averaged over a 12° x 12° sub-grid of the GFS data field that is centered on the algorithm sub-region.

Mid-level moisture levels may influence the development of tropical cyclones by modulating convective activity. Dry air at middle levels can have a negative impact on convective activity by reducing updraft buoyancy via entrainment and decreasing the precipitation efficiency within developing systems (Ruprecht and Gray 1974; Gray 1975; Bister and Emanuel 1997). As described by Moody et al. (1999), the GOES water vapor imagery in cloud-free regions is sensitive to the relative humidity in the mid- to upper troposphere. Hence, the average brightness temperature (BTWARM) within each sub-region is calculated after removing all pixels colder than -40 °C and is used as an approximation of mid-level moisture.

Observational studies have found that deep convective bursts are often a precursor to TC formation (Gentry et al. 1970; McBride and Zehr 1981; Zehr 1992; Molinari 2004). In order to include this environmental signature in our algorithm, a parameter to quantify deep convective activity was developed. This parameter (PCCOLD) uses water vapor (6.7 μm) brightness temperatures and determines the percent of pixels within each sub-region that are colder than -40°C.

Monthly climatological TC formation probabilities were calculated by adding up the total number of TC formations in each sub-region for each month over the 1949-2005 best track dataset and then dividing by the number of years (57). These monthly values were then divided by the total number of days in each month to obtain an average daily climatological TC formation probability for each month. Finally, the 24-hour climatological probability parameter (CPROB) was calculated by linearly interpolating between these average daily probabilities.

Australian Bureau of Meteorology, 2007: Forecast Verification – Issues, Methods, and FAQs. http://www.bom.gov.au/bmrc/wefor/staff/eee/verif/verif_web_page.html.

Beven, J.L., 1999: The boguscane – A serious problem with the NCEP medium range forecast model in the Tropics. Preprints, 23rd Conf. On Hurr. and Trop. Meteor., Dallas, TX, Amer. Meteor. Soc., 845-848.

Bister, M., and K.A. Emanuel, 1997: The Genesis of Hurricane Guillermo: TEXMEX Analyses and a Modeling Study. Mon. Wea. Rev., 125, 2662–2682. Briegel, L. M. and W. M. Frank, 1997: Large-scale forcing of tropical cyclogenesis in the Western North Pacific. Mon. Wea. Rev., 125, 1397-1413.

Bracken, W.E., and L.F. Bosart, 2000: The Role of Synoptic-Scale Flow during Tropical Cyclogenesis over the North Atlantic Ocean. Mon. Wea. Rev., 128, 353–376.

Briegel, L.M., and W.M. Frank, 1997: Large-Scale Influences on Tropical Cyclogenesis in the Western North Pacific. Mon. Wea. Rev., 125, 1397–1413.

Davis, C., C. Snyder and A.C. Didlake, Jr., 2007: A Vortex-based Perspective of Eastern Pacific Tropical Cyclone Formation. Mon. Wea. Rev., submitted.

DeMaria, M. M. Mainelli, L.K. Shay, J.A. Knaff and J. Kaplan, 2005: Further Improvements in the Statistical Hurricane Intensity Prediction Scheme (SHIPS). Wea. Forecasting, 20, 531-543.

____, J.A. Knaff, and B.H. Connell, 2001: A Tropical Cyclone Genesis Parameter for the Tropical Atlantic. Wea. Forecasting, 16, 219–233.

Elsberry, R. L., W. M. Frank, G. J. Holland, J. D. Jarrell, and R. L. Southern, 1987: A Global View of Tropical Cyclones. University of Chicago Press, 185 pp.

Emanuel, K. A., 1989: The finite amplitude nature of tropical cyclogenesis. J. Atmos. Sci., 46, 3431–3456.

____ and Nolan, D. S. 2004. Tropical cyclone activity and global climate. Proc. of 26th Conference on Hurricanes and Tropical Meteorology, American Meteorological Society, pp. 240–241.

Frank, N. L., and G. Clark, 1980: Atlantic tropical systems of 1979. Mon. Wea. Rev., 108, 966–972.

____, W.M., and P.E. Roundy, 2006: The Role of Tropical Waves in Tropical Cyclogenesis. Mon. Wea. Rev., 134, 2397–2417.

Gentry, R.C., T.T. Fujita, and R.C. Sheets, 1970: Aircraft, Spacecraft, Satellite and Radar Observations of Hurricane Gladys, 1968. J. Appl. Meteor., 9, 837–850.

Gray, W. M., 1968: Global view of the origin of tropical disturbances and storms. Mon. Wea. Rev., 96, 669–700.

____, 1975: Tropical Cyclone Genesis in the Western North Pacific. Tech. Paper No. 16-75, 66 pp. [Available from the Dept. of Atmospheric Sciences, Colorado State University, Ft. Collins, CO 80523].

____, 1979: Hurricanes: their formation, structure and likely role in the tropical circulation. In: Meteorology over the Tropical Oceans. Roy. Meteor. Soc., pp. 155–218

____, 1984: Atlantic seasonal hurricane frequency. Part I: El Nino and 30 mb quasi-biennial oscillation influences. Mon. Wea. Rev., 112, 1649–1668.

Harr, P. A., cited 2007: Joint hurricane testbed: Final report, Objective and automated assessment of operational global forecast predictions of tropical cyclone formation and life cycle. [Available online at http://www.nhc.noaa.gov/jht/final_rep/ENharr_JHTfinalreport_03-05.pdf]

Hendricks, E.A., M.T. Montgomery, and C.A. Davis, 2004: The Role of “Vortical” Hot Towers in the Formation of Tropical Cyclone Diana (1984). J. Atmos. Sci., 61, 1209–1232.

____, and M.T. Montgomery, 2006: Rapid Scan Views of Convectively Generated Mesovortices in Sheared Tropical Cyclone Gustav (2002). Wea. Forecasting, 21, 1041–1050.

Hennon, C.C., and J.S. Hobgood, 2003: Forecasting Tropical Cyclogenesis over the Atlantic Basin Using Large-Scale Data. Mon. Wea. Rev., 131, 2927–2940.

Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc., 77, 437–471.

Knaff, J.A., T.A. Cram, A.B. Schumacher, J.P. Kossin and M. DeMaria, 2008: Objective identification of annular hurricanes. Wea. Forecasting, 23, in press.

Leipper, D.F., 1967: Observed Ocean Conditions and Hurricane Hilda, 1964. J. Atmos. Sci., 24, 182–186.

Malkus, J.S. and H. Riehl, 1960: On the dynamics and energy transformation in steady-state hurricanes. Tellus, 12, 1-20.

Maloney, E.D., and D.L. Hartmann, 2001: The Madden–Julian Oscillation, Barotropic Dynamics, and North Pacific Tropical Cyclone Formation. Part I: Observations. J. Atmos. Sci., 58, 2545–2558.

Mason, S. J. and N. E. Graham, 1999: Conditional probabilities, relative operating characteristics, and relative operating levels. Wea. Forecasting, 14, 713-725.

McBride, J. L., and R. M. Zehr, 1981: Observational analysis of tropical cyclone formation. Part II: Comparison of non-developing versus developing systems. J. Atmos. Sci., 38, 1132–1151.

____, 1995: Tropical cyclone formation. Global Perspectives on Tropical Cyclones, WMO/TD No. 693, Rep. TCP-38, World Meteorological Organization, 63–105.

Molinari, J., D. Vollaro, S. Skubis, and M. Dickinson, 2000: Origins and Mechanisms of Eastern Pacific Tropical Cyclogenesis: A Case Study. Mon. Wea. Rev., 128, 125–139.

____, and D. Vollaro, 2000: Planetary- and Synoptic-Scale Influences on Eastern Pacific Tropical Cyclogenesis. Mon. Wea. Rev., 128, 3296–3307.

____, D. Vollaro, and K.L. Corbosiero, 2004: Tropical Cyclone Formation in a Sheared Environment: A Case Study. J. Atmos. Sci., 61, 2493–2509.

____, K. Lombardo, and D. Vollaro, 2007: Tropical Cyclogenesis within an Equatorial Rossby Wave Packet. J. Atmos. Sci., 64, 1301–1317.

Montgomery, M.T., M.E. Nicholls, T.A. Cram, and A.B. Saunders, 2006: A Vortical Hot Tower Route to Tropical Cyclogenesis. J. Atmos. Sci., 63, 355–386.

Moody, J. L., A. J. Wimmers, and J. C. Davenport, 1999: Remotely sensed specific humidity: Development of a derived product from the GOES Imager channel 3. Geophys. Res. Lett., 26, 59–62.

Murphy, A.H., 1973: A New Vector Partition of the Probability Score. J. Appl. Meteor., 12, 595–600.

____, 1991: Probabilities, Odds, and Forecasts of Rare Events. Wea. Forecasting, 6, 302–307.

Ooyama, K. V., 1990: A thermodynamic foundation for modeling the moist atmosphere. J. Atmos. Sci., 47, 2580–2593.

Palmén, E.H., 1948: On the formation and structure of tropical cyclones. Geophysics, 3, 26-38.

Pasch, R. J., P. A. Harr, L. A. Avila, J-G Jiing, and G. Eliott, 2006: An evaluation and comparison of predictions of tropical cyclone cyclogenesis by three global forecast models. Proc. of the AMS 27^th Conference on Hurricanes and Tropical Meteorology, American Meteorological Society, CD-ROM. [Available online at http://ams.confex.com/ams/pdfpapers/108725.pdf ]

Perrone, T.J., and P.R. Lowe, 1986: A Statistically Derived Prediction Procedure for Tropical Storm Formation. Mon. Wea. Rev., 114, 165–177.

Riehl, H., 1948: On the formation of typhoons. J. Meteor., 5, 247–264.

Ruprecht, E. and W.M. Gray, 1974: Analysis of satellite-observed tropical cloud clusters. Tech. Paper No. 219, 91 pp. [Available from the Dept. of Atmospheric Sciences, Colorado State University, Ft. Collins, CO 80523].

Sampson, C. R. and A. J. Schrader, 2000: The automated tropical cyclone forecasting system (Version 3.2). Bull. Amer. Met. Soc., 81, 1231-1240.

Shapiro, L. J., and S. B. Goldenberg, 1998: Atlantic sea surface temperatures and tropical cyclone formation. J. Climate, 11, 578–590.

Simpson, J., J.B. Halverson, B.S. Ferrier, W.A. Petersen, R.H. Simpson, R. Blakeslee, and S.L. Durden, 1998: On the role of “hot towers” in tropical cyclone formation. Meteor. Atmos. Phys., 67, 15-35.

Stephenson, D., 2004: Skill Measures for Forecasts of Extreme Rare Events, Climate Extremes and Risk Reduction Conference, Lighthill Institute of Mathematical Science, London, England, November 30, 2004. http://www.ucl.ac.uk/lims/events/LIMSConf30Nov04.htm

Wilks, D. S., 2006: Statistical Methods in the Atmospheric Sciences, Second Edition, Academic Press, 467 pp.

Zehr, R. M., 1992: Tropical cyclogenesis in the western north Pacific. NOAA Tech. Rep. NESDIS 61, 181 pp. [Available from NOAA/National Environmental Satellite, Data, and Information Service, Washington, DC 20233.]

TABLE1. The criteria used to eliminate data points from the dataset during the screening step. Note that climatological probability (CPROB) was not used for screening because it is derived from a limited Best Track dataset (1949-2005). In the third column, NCEP indicates the data source is the NCEP GFS analyses and SAT WV indicates the data source is satellite water vapor (6.7 μm) imagery.

TABLE 2. The number of TCG and NTCG cases observed versus the number of TCG and NTCG cases predicted by each consecutive step of the algorithm. The resultant hit rates (HR) and false alarm rates (FAR) are also shown.

Table 3: Standardized discriminant coefficients for each parameter in each basin, derived from linear discriminant analysis after screening.

FIG. 2. All tropical cyclone formation points from 1949-2005 best tracks.